The box reads "Ask Laureo, or describe what you want done", or names the record you are looking at when the panel is open on one. Send is the square button to its right. While a run is going, that same square becomes Stop.

Keyboard

• By default, Enter sends and Shift+Enter starts a new line.
• If you set Send messages with to Cmd+Enter in chat settings, plain Enter starts a new line instead and only Cmd or Ctrl with Enter sends.
• Escape stops a run that is generating. If a menu is open in the box, it closes that menu first.
• Typing in a language that uses a composition step, such as Japanese or Chinese, never sends on the key that accepts a candidate.

Attaching things

Add context is the plus control inside the box, at its bottom right. It offers three things: Upload file, CRM file for something already stored in Laureo, and Record for a contact, company, deal, quote, product, project, campaign, or ticket. Search there only ever shows records you can currently access.

What you attach appears as chips above the box, each with its own remove control. Laureo takes a snapshot of them when your message is accepted, so removing a chip later changes future messages without changing the file or record it came from.

Mentioning a record with @

Type @ at the start of a message or after a space, then at least two characters, and a short list of matching records appears. An email address you type never triggers it. Arrow keys move through the list, Enter or Tab inserts, and Escape dismisses it. Enter never sends while that list is open. See Mentioning a Record below for which types also attach themselves.

Slash commands

Type / at the start of the box to open the command list. Arrow keys move, Tab completes, and Escape clears the slash. Full list in Slash Commands.

Length

Messages are capped at 5,000 characters. Past 4,000 a quiet counter appears inside the box reading, for example, 4,320 of 5,000. It turns amber as you approach the cap and red past it, and the line under the box says why Send is unavailable: "Message too long: trim to 5,000 characters."

Sending while something else is happening

The box is never locked while Laureo is working, and never locked by a proposal waiting on you. You can always type.

• A run is going: Send queues your message behind it, and a strip above the box reads "{n} waiting to send" with Edit, Up, Down and Remove on each row. If instead you want the new message to replace the work in flight, use Stop current task and send, which appears under the box while a run is active.
• A proposal is waiting: a muted line reads "Waiting for your approval above." You can still type. Sending asks first, with "Send this instead?" and "The action waiting for approval will be cancelled.", offering Send or Keep waiting.

Send completes only once Laureo has durably accepted the command. After that, closing the panel or the browser does not cancel it.

Example questions:

• "Show me my at-risk deals"
• "What tasks are overdue?"
• "Find companies in the healthcare industry"
• "Summarize my pipeline health"
• "Create a follow-up task for Acme Corp due next Monday"
• "Help me prioritize my deals this week"
• "Draft an email to Sarah about the renewal"

The Add context menu has three choices: upload a new file, select a file already stored in Laureo, or select a record such as a contact, deal, quote, campaign, product, or booking. Record references stay structured and are not copied into fake files.

• Text formats: plain text, Markdown, CSV, and JSON.
• Office formats: XLSX, DOCX, and PPTX. Macros, embedded objects, password protection, external relationships, and unsafe archives are refused.
• PDF: support is narrow on purpose. A PDF is read when its text is selectable, it uses the standard Helvetica, Times, or Courier fonts, and it was saved in one piece without a password. Everyday files fall outside that: one that was edited or signed electronically and saved as an update on top of the original, one from a newer editor that writes a compact cross-reference layout, and one whose fonts were embedded as partial copies, which is what exporting or printing to PDF usually produces. Image-only scans and files with active content are refused too. Each refusal shows a file error naming what is supported, and no text is ever guessed. Attaching the original as DOCX, XLSX, or text is the reliable path.
• Images: on the Business plan and above, paste, drop, or attach a PNG, JPEG, WebP, or GIF and Laureo reads what is in it and answers from it. A screenshot of an error, a signed page, or a spreadsheet a colleague sent you all work. Any other image format is refused, and so is every image on a plan that does not carry the capability, or in an organization where it has not been switched on. The refusal happens before the file is stored, and it names which of those two reasons applies, so a file is never accepted and then quietly ignored.

Each message can include up to 10 references. File size and storage limits follow your plan. Extracted context is capped per reference and across the message, so a very long document may be shortened to the portion that fits safely.

The AI assistant is designed to be a business advisor grounded in your CRM data. It will help with sales strategy, pipeline planning, email composition, and workflow optimization.

If you ask about topics outside its scope (code, medical advice, general trivia), it will redirect you back to CRM-related help. For competitor questions, it will only assist with migrating data from another CRM into Laureo.

While Laureo is working there is exactly one block describing that run, and it sits in the conversation itself rather than in a footer bar or a floating pill. Its top line holds three things:

• What is happening right now, as a short verb and object such as "Searching deals" or "Saving quote". The line is replaced as the run moves on, so a long run does not bury the conversation in narration. Before there is anything to name, it reads Working on it; while Laureo is still reattaching to a run that is already going, it reads Connecting to the running task.
• How long it has been going, in whole seconds.
• Stop. Pressing it changes the label to Stopping immediately, before the server has answered, so the button never lies about what you just did.

On a narrow screen the line wraps: the activity and the timer on one line, Stop on the next.

The steps underneath

Each step Laureo takes is listed under that line. Up to three of the most recent are shown outright, two on a small phone. From four onward the rest are behind Show all {n} steps. The list never opens itself and never scrolls itself while you are reading. The same thing done several times in a row collapses into one row, such as "Searching deals, 3 times", rather than three identical lines.

A written plan appears only when a run declares five or more steps in advance, and even then it arrives collapsed as Plan: {n} steps. Below that, the step list is the record. Most runs show no plan at all, which is deliberate.

The thinking line

When Laureo is working out what to do and has produced neither text nor a step yet, the same line reads Thinking for {n}s and counts up. That is all it ever shows. Private reasoning is never displayed, summarized, stored, or exported, not even as a length or a first line.

When the answer starts

The moment the answer starts arriving, the working block's top line disappears and the answer takes its place. You never see both at once. In its place, under the finished answer, a quiet Did {n} steps control keeps the full step list available for as long as the conversation exists. Traces do not vanish when a run ends.

Beside the activity line, a short word names the kind of work the run was routed to, such as a short answer or an agentic run. It is read from the run's own saved record rather than guessed, so if that record has not been read back yet nothing is shown instead of a guess. It never names a provider or a model.

A proposal is a card in the conversation, never a pop-up over it, so the rest of the app stays usable while you read it. It is laid out top to bottom:

• The action, as a verb and its object: Create task, Update deal, Send quote.
• Why it is asking, in one short sentence, when Laureo has a reason worth stating. If it has none, the line is simply absent rather than filled with something invented.
• The fields it will write, one row each: Title, Due, Assignee, Priority, and so on. Six rows are shown, then Show all {n} fields. Values are shown as words rather than raw data.
• Approve, Edit, and Deny.

The lines above the buttons

Up to three short muted lines can appear between the fields and the buttons. Each is absent unless it is true, so a card is never padded out with reassurance:

• Where the content came from, when the run read something from outside the CRM before proposing this write, such as "This proposal used content from an email." or "This proposal used content from an attachment." Content that arrived from outside always earns you a look, even in automatic mode.
• Whether it can be taken back: "You can undo this afterwards.", "You can correct this afterwards, but not undo it.", or "This cannot be undone." An action that changes no record says nothing here.
• How to put it back, when there is a specific route to reversing it.

Approve, edit, deny

• Approve records your decision and runs the effect. The button reads Approving while it is in flight.
• Edit turns the field rows into inputs so you can change a value in place, and Approve becomes Save and approve. Cancel puts the original values back. Edit appears only where there is genuinely something editable; where there is not, a line under the buttons reads "This action cannot be edited here. Deny it and say what to change."
• Deny opens a single optional field, "What should it do instead? (optional)". Leave it blank and the turn ends. Type a correction and the run keeps going with your correction, so one wrong field does not cost you the whole request. Pressing Escape opens that field, and pressing it again denies.

What the card becomes afterwards

• Done: the card collapses to one receipt line with a tick, such as "Task created." and an Open link straight to the record. Following it keeps the conversation with you.
• Still finishing: "Saved. Finishing this in the background." The card does not linger with dead buttons.
• Denied: the card is replaced by a single line, "You denied this. Nothing was changed."
• Expired: the action, one line reading "This proposal expired.", and one Ask again control. Ask again re-sends your original request. It never silently re-approves anything.

Batches keep a checkbox per item and read Approve selected, Approve all, or Deny all. Some batches are all or nothing and say so. Risk is stated once, in words, in the why line, rather than as a badge beside every row.

An approval is bound to the run and conversation it came from, and cannot authorize a different action later. Before the effect actually runs, Laureo rechecks your current membership, role, plan, scope, the record version, and the recipients. A double press, a second tab, or a retry cannot make the same change twice.

Automatic mode may let reviewed, reversible preparation run without a card. It never skips a proposal for sending, publishing, spending, deleting, or changing permissions. Which individual actions may run without asking is a separate organization setting on the Business plan and above; see Approval settings.

A proposal appears in exactly two places: once as a card in the conversation it belongs to, and once as a row in the assistant chip. There is no footer bar and no floating pill duplicating it.

On a wide screen the chip sits in the top right of the app bar, beside the notification bell. It reads Needs approval with a count when something is genuinely blocked on you, and the quieter Working with a count when runs are simply in progress. With nothing pending and nothing running it does not render at all.

On a narrow screen, where there is no app bar, the same chip floats in the top right corner instead, and its list opens as a bottom sheet. It hides itself while you are on an assistant surface, because the conversation in front of you already shows the card.

Opening it lists each run with what it is waiting for and the conversation it belongs to. From a wide screen you can Approve or Deny straight from the row without opening the conversation, or press Open to go to it. If a decision cannot be sent, the panel says "Could not send that decision. Open the task to try again." rather than pretending it landed.

In the conversation list, a conversation whose run needs approval carries a small dot before its title.

When a goal spans several domains, has materially different strategies, or contains meaningful approval checkpoints, Laureo presents one execution plan. It shows the objective, deliverables, defaults, affected records, ordered steps, and future approval checkpoints.

Approve plan authorizes that displayed strategy and its safe, reversible preparation. It does not waive a later exact approval to send, publish, spend, delete, change permissions, or create another irreversible effect.

• Material missing information appears in one grouped clarification card rather than a series of single-field questions.
• Safe defaults and their source appear in the plan before you approve it.
• Every irreversible checkpoint receives its own exact preview at the moment it is ready to run.

A plan can remain waiting while you are away. When you approve it, Laureo rechecks that it still belongs to your run, conversation, organization, and current authority. A material edit creates a new plan preview.

After an eligible plan is approved, Create Agent draft opens an unpublished Agent template with the objective, ordered steps, scope placeholders, and currently reviewed read-only capability suggestions. The source conversation stays linked for review. The template does not inherit the plan approval, recipients, attachments, publication permission, live trigger, or run-as authority. You must choose the Agent record type and trigger, and normal admin grant rules still apply before publishing.

When Laureo needs more than one detail before it can act, it shows a clarification card: a short form in the conversation with one row per open question. It is deliberately not a series of questions, because answering five things one message at a time is five round trips for one decision.

• Sensible answers arrive already filled in. A field the product can work out for itself starts carrying that value, with a quiet word beside the label saying it is still an assumption. The word disappears the moment you change the value, because at that point it is your answer.
• It never asks about something the product already decides. Currency, quantity, item type, a quote title, an expiry date and payment terms all resolve without you, so they are filtered out of the question list rather than asked.
• Continue sends everything at once as a single message. The button reads Saving answers while it is in flight.
• Skip: let the assistant decide is there whenever something is left that the assistant may actually decide, so a question asked on a wrong guess is never a dead end. It hands over the terms you did not answer, keeps the ones you did exactly as you typed them, and hands over nothing else. Every step that leaves the product or cannot be undone still meets its own approval card afterwards. When every question still open is one only you can answer, Skip is withdrawn rather than shown as a button that would move nothing, and the card is left as a question only you can answer, with Continue.
• A verification hand-off is one of those cards. When a step needs fresh multi-factor verification before it may run, the task pauses and the card carries a single Open Security verification link instead of a field to fill in. Open it, verify, come back and press Continue, and the task picks up where it stopped. The link authorizes nothing on its own, and Skip is withdrawn there because no one else can verify for you.

Once you submit, the form is replaced by a one-line recap of the terms the run is now working from, separating the two honestly: "You answered ..." for what you typed, and "Assumed ..." for what was taken as read. The recap stays in the transcript, so what the run was told is still legible later.

Ask take me to the vendor list or where do I set up booking pages and Laureo answers from the real list of pages you can open, filtered exactly the way your sidebar and command palette are filtered. Accept and the app navigates, carrying the conversation with you. Moving you to a page creates nothing and sends nothing, so it raises no approval card; Back is the whole undo.

Ask how does the scheduler work and Laureo searches this help centre, reads the article that answers you, and replies in the product's own words with a link to the article it used. Follow the link when the answer looks thin; the article is always longer than the summary.

On the Ultra plan, Laureo can build a cold outreach sequence with you in chat. Ask for what you want: "draft me a three-step sequence for agency founders": and it writes the whole sequence in one go: the first email with its subject line, then each threaded follow-up with how long to wait and whether to send it if the prospect already replied. The sequence is created as a draft, which emails nobody.

From there the flow is four steps, and only the last one sends anything:

• Draft the sequence: an approval card shows every step so you can read the copy before it is saved.
• Set the audience: attach one or more of your lead lists. Still no email.
• Check it: ask Laureo whether it is ready to launch and it reports what the Launch button would: whether the steps are structured correctly, whether you have a verified sending mailbox that is not paused, and how many eligible leads would actually receive it.
• Launch: the only step that sends. It always stops and asks, even in automatic mode.

Laureo can only see and change the sequences and lead lists you could see and change yourself. If your role gives you access to your own outreach rather than the whole team's, a teammate's sequence is simply not found: the same answer you get in the Cold Outreach pages.

Alongside contacts, deals, email and the calendar, Laureo can now answer questions about the parts of the business that used to need a trip to a settings page:

• The Recycle Bin: "what did I delete this week", "is that company still recoverable". It can list deleted companies, contacts, deals and tickets for the 30 days before they are purged.
• Duplicates: "do we already have Acme" before you create a second record. It reports what matched and why: an exact account number, an exact name, a similar name, or a contact email already on file.
• Imports: "did my import finish", "how many rows failed". It can read the import history or one batch's progress and error count.
• Purchase and sales orders: "what are we waiting on from suppliers", "which orders are unfulfilled", and the line-by-line detail of any one of them.
• Enrichment: whether it is set up at all and what a lookup would cost, so it can tell you before offering rather than after failing.
• Lead capture: which forms are live, and who has submitted them recently. Managers and admins only, matching the Lead Capture pages.

Every one of these is read-only, and every one is limited to exactly what you could see yourself. If your role gives you your own sales orders rather than the whole team's, that is what Laureo reads. If you cannot open the Lead Capture pages, Laureo will not read them for you either.

Laureo can also see the automations you have running, the booking pages people book you through, and the notifications waiting on you. As questions, these are read-only:

• Automations: "which automations are on", "when did the lead-routing rule last run". It lists them with their status and last-run time, and can open a single run to tell you which steps executed and what the error was if it failed. Cold outreach sequences are a separate thing and are not included here.
• Booking pages: which pages are live, how long each meeting is, and who hosts it.
• Notifications: what is currently waiting on you.

Four of these are actions rather than questions. Each one shows you exactly what it is about to do and waits for you to confirm, every time, in every mode:

• Running an automation: starts the whole sequence, which can email people and change records as it goes.
• Pausing or resuming a booking page: pausing stops new bookings from the moment you confirm. Bookings already in the diary are untouched, and turning the page back on restores it.
• Marking meeting attendance: records attended, no-show, or clears the mark. It is the same mark you would set by hand, and it closes the matching "mark this meeting" item in My Work.
• Posting to team chat: posts into an internal conversation you are already part of, and your teammates are notified. This is internal chat only; to reach someone outside the company, ask for an email instead.

Alongside the read-only answers above, Laureo can make two changes here. Both show you what will happen and wait for you to confirm, every time, in every mode:

• Restore from the Recycle Bin: "put the Acme record back". Laureo finds the deleted record, shows you which one it means, and restores it once you confirm. Restoring needs the same permission as deleting, so if you could not have deleted it, Laureo cannot put it back for you.
• Edit a draft campaign: change its name, its subject line, or which contact group it goes to. Only drafts: once a campaign has been sent or scheduled, Laureo will say so rather than change it. Editing never sends anything.

Both go through exactly the same checks as doing it by hand: the same permissions, the same ownership rules, the same audit trail. Restoring a record shows up in the audit log under your name, and the record comes back on your teammates' screens straight away.

A handful of things Laureo can do are permanent, cost money, or reach someone outside your organization. Every one of them shows you exactly what will happen and waits for you to confirm: including in automatic mode, which cannot skip them:

• Merging duplicates: one record is folded into the other and stops existing. There is no un-merge, and the preview names which record disappears.
• Exporting records: prepares a downloadable file of a whole record type. Managers and admins only. Laureo never reads the exported rows back into the chat.
• Adjusting stock: writes a permanent entry to the stock ledger. Ledger entries cannot be edited or removed; a mistake is corrected with a second adjustment.
• Sending a purchase order: emails an outside vendor and marks the order sent.
• Sending a payment reminder: a real email to a customer about an overdue invoice. Ask for a draft instead if you only want to see the wording.
• Looking up details from an outside provider: spends enrichment credits on the attempt whether or not anything is found.
• Posting a message to a teammate: lands in their inbox and notifies them straight away. There is no recall.
• Running an automation: starts the whole sequence of steps, which can email people and change records as it goes. Undoing a run means undoing each step it took.
• Requesting money out of a trust account: see below.

All of these need the same permission you would need to do them by hand, and several need more: sending a payment reminder needs both collections access and the right to send email as yourself, and requesting a trust disbursement needs organization-wide trust permission rather than access to your own records.

Laureo can create tasks, update existing ones, and optionally sync a task to your connected Google or Outlook calendar: all from chat.

• Create and sync in one step: "Create a task to call Acme due Friday and put it on my calendar" creates the task with a due date and, when you have a calendar connected, a matching calendar event.
• Sync an existing task: "Sync the Acme follow-up to my calendar" updates the task so it appears on your calendar, with no need to recreate it. This is the same as toggling calendar sync on the task itself.
• Update task details: "Move the Acme task to next Tuesday" or "Reassign the onboarding task to Sam" edits the task in place. Reassignment respects the same permissions as editing the task by hand.

When you ask Laureo about your schedule ("What's on my calendar this week?"), tasks you've synced to your calendar show up alongside your meetings and activities, sorted by date. Tasks you haven't synced stay out of that view.

After a completed text turn, feedback and regenerate controls appear in a full-width utility row below approval and artifact actions. They stay hidden while a decision is pending so they never compete with the primary action.

Once you rate a message, the selected icon highlights (green for positive, red for negative) and cannot be changed.

When an answer was grounded in your records, a Sources line appears under it with the number of records in brackets. It starts closed and never opens itself, so a long answer does not push the message box around while you are reading. Open it to see each record by name, with a short excerpt where one is available.

The list is built from what the run actually read, on the server, so it is a record of the answer rather than a claim about it. An answer that read nothing shows no Sources line at all instead of an empty one.

Selecting a source opens that record and keeps this conversation with you in the desktop dock or mobile sheet, so checking a source never costs you the thread. A record whose type has no page of its own is listed as plain text rather than as a link that would go nowhere.

Small reference numbers in the answer. Where a sentence rests on one particular record, a superscript number follows it. Pressing it opens the Sources list and takes you to that row, so you can tell which of eight cited deals produced the number in front of you. A number the list cannot resolve stays as plain text rather than becoming a control that goes nowhere.

Counts are links. When an answer says something like "7 deals are stalled", that count becomes a link to the filtered list it came from. Both the number and the destination come from the search Laureo actually ran, so a count can never link to a set it invented. Only the first mention of each count is linked, so prose does not turn into a wall of links.

The answer arrives as an ordinary message that grows as Laureo commits it. There is no blinking cursor and no shimmer. If your connection drops, reopen the conversation and the saved events replay from the last one you saw. The browser stream is a view of the work, never the work itself.

Stop is saved on the server. It blocks anything further from being claimed and asks the current run to abort. Text and changes already committed stay in the conversation, honestly recorded, rather than being described as cancelled.

System notes

Facts about the run itself arrive as one short centred line, not as a message from Laureo. They carry no timestamp, no copy control and no feedback buttons, because there is nothing to rate. The wording is fixed:

• You pressed Stop: "You stopped this run."
• Cancelled before anything ran: "This run was cancelled. Nothing was changed."
• Cancelled after steps ran: "This run was cancelled. The steps above already ran."
• Failed before anything ran: "This run failed. Nothing was changed."
• Failed partway: "This run failed partway. The steps above already ran, and nothing after them did."
• You denied a proposal: "You denied this. Nothing was changed."
• A proposal ran out of time: "This proposal expired."
• A big job was accepted: "This is a large run. It will pause for approval before changing anything."
• You saved or removed a preference: "Saved to your preferences." or "Removed from your preferences.", quoting what you saved, with Forget this or Undo beside it and an Open link to the full list.

Time separators

Messages no longer each wear a timestamp. Instead, a small centred line marks a real gap: when more than fifteen minutes passed or the day changed, you see Today 4:12 pm, Yesterday 9:40 am, 3 Sep, 4:12 pm, or the year as well for anything older. The exact time of a single message is still there on hover and for screen readers.

Follow-up chips

After a finished answer, up to three one-tap follow-ups appear under it. They stay away while a run is going and while a proposal is waiting on you, so they never compete with a decision. See Voice & Follow-up Chips.

The run summary

A run that did several things closes with a short ledger under the answer: "Did 6 steps: 4 done, 1 skipped, 1 failed", then one line per step. Each step that landed carries its own Open link to what it created or changed; a skipped or failed step says why. The lists are read back from what the run actually recorded, so it is a record rather than a closing sentence Laureo composed. A run with nothing to report shows no summary rather than an empty one.

Every conversation is saved as you go. On the full page the list is a rail beside the chat; in the side panel and the home assistant, the Conversation history control opens the same list over the chat. Rows are grouped, in this order: Pinned, Today, Yesterday, Past 7 days, Past 30 days, Older. Days are counted in your own time zone. A pinned conversation sits under Pinned rather than under its date.

Each row shows the title and how long ago it was used. A conversation whose run is waiting on your approval carries a small dot before its title. A conversation with no title yet reads New chat, the same phrase as the button that creates one.

On a desktop screen you can widen the side panel by dragging its left edge, or by focusing that edge and using the left and right arrow keys. The width is remembered on that device.

Titles

A new conversation is named from your first message immediately, so the row never sits blank. After the first answer, Laureo replaces it with a short generated title. Once that generated title exists, renaming the conversation yourself is the last word and nothing renames it again. A rename made before the first answer arrives can still be replaced by that one generated title.

The row menu

Each row on the full page carries a ⋯ menu, reachable by touch as well as by hover:

• Pin to top and Unpin, which move it into or out of the Pinned group
• Rename, edited in place on the row
• Archive and Unarchive, which move it between the Conversations and Archived views without deleting anything
• Export, which builds a downloadable record of the conversation. It is prepared in the background: if it is not ready within a few seconds you see "Your export is still being prepared." and it downloads on its own when it is. A failure says "Could not export this conversation." and nothing is changed.
• Delete, which asks first: "Delete this conversation?" and "... and all of its messages will be removed. This cannot be undone." Laureo fences any active work and cancels pending proposals before it goes.

The history list inside the side panel is for choosing a conversation and starting a new one. Rename, archive, export and delete live on the full page.

Search

The box reads Search conversations and searches the whole of your history on the server, not just the rows already loaded. It matches four things, in this order of preference: the title, the text of messages, the labels on receipts, and the names of attachments. A hit shows a short excerpt under the title, and opening it takes you to the message that matched rather than to the top of the conversation. With no matches it says "No conversations match 'your term'."

Linking to a conversation

A conversation lives at /ai/<conversation id>, and that address opens it for you alone. Conversations are private to the person who started them, so sending the address to a colleague only shows them a conversation that is not available. Notifications about a finished run link straight to the answer inside it. Switching conversations on the full page keeps the address in step, so Back and Forward work the way you expect.

Other list controls

• New chat starts a fresh conversation. Its row appears at the top of Today straight away.
• Load older conversations at the bottom of the list fetches the next page.
• An empty list says "Your conversations will appear here.", and a list that could not load says "Could not load your conversations." with a Retry control.

The assistant keeps your recent exchanges word for word. Once a chat runs past that, it starts condensing the oldest part into a summary so the conversation can keep going. Two things tell you where you are:

• A line above the message box, once a chat is long enough for condensing to be near. It says whether earlier messages are still intact or have already been condensed, and offers one "New chat" link. If your next message looks like a change of subject, the same line says so instead.
• A notice in the conversation at the moment condensing happens, with the number of messages it covered. Expand "What the assistant remembers" to read the summary it kept.

Details from the very beginning of a condensed chat may be missing, so if you are moving to an unrelated question, a new chat gives the assistant a clean, focused context to answer from.

• "You are offline. Messages will fail until your connection is back." Your device cannot send a new message until connectivity returns. Work already accepted carries on regardless.
• "Live updates paused. Refresh to see new messages." The live view lost its connection. Nothing is lost: reconnecting replays the saved events in order.

Everything else Laureo can say when something goes wrong is listed in When something goes wrong.

Read-only and text-only turns can be regenerated as preserved alternatives. The earlier answer remains in history so audit context is not destroyed.

After a product, quote, send, or any other CRM or external effect commits, destructive edit-and-resend and regenerate are unavailable. Receipts remain immutable. Continue with a corrective request, or use an offered undo when supported.

Type @ followed by at least two letters to search your records. Pick one with Enter, Tab or a click and its name drops into your message. Arrow keys move through the list; Escape closes it without losing what you typed. Each row shows the record name, a short detail, and what kind of record it is.

Mentioning a company, person, deal or quote also attaches that record as context, exactly as picking it from Add context would, and it counts toward the same limit of 10 references on a message. A ticket mention is text in your message and nothing more, so use Add context if you want the ticket itself attached.

Naming records also narrows the answer. Once you have pointed a question at particular records, Laureo grounds the reply in them, and the Sources list under the answer can only name records inside what you pointed at. It keeps the tools it would normally use, because a question about a deal still has to read that deal's own related rows to be answered at all.

The list only shows records you already have permission to see, and never shows anything in the recycle bin or an archived quote.

Open the three-dot menu on any conversation in the sidebar and choose Pin to top. Pinned chats gather in their own Pinned group above the date groups, so a thread you pinned months ago does not sink as you start new ones. Choose Unpin from the same menu to return it to its place by date.

Pins are yours alone. They change nothing about the conversation itself and nobody else sees them.

Above the message box, Model: Auto opens a short list. Auto is the default and is usually right: it picks per message, matching a quick lookup to a fast model and a multi step request to a stronger one. Choosing a specific option pins it for that one conversation only.

• Fast: quickest answers, best for lookups and short questions.
• Standard: a balanced default for everyday work.
• Advanced: for nuanced, multi step requests. Costs more per message.

The names match the ones on the agent model picker in Automations, and they describe the class of model rather than a specific one, so your choice keeps working as models change. The control is hidden until a chat has its first message, and is unavailable while an answer is being written.

Hover over any message to reveal a copy button. Click it to copy the message text to your clipboard. This works for both your messages and AI responses.

On a narrow screen the assistant fills the screen rather than floating over it, and every control is sized for a fingertip.

• The conversation list on the full page slides in from the left. Close it by tapping outside it, by swiping it away, or with the close control in its header. A gesture is never the only way out.
• Menus open as bottom sheets. The permission mode menu, the approvals list, the work-in-other-chats list and every confirmation come up from the bottom edge, keep your place on the page behind them, close on a tap outside or on the back gesture, and sit clear of the home indicator and any camera cutout.
• The approvals chip lives in the top right corner beside the notification bell, since the bottom of the screen belongs to the navigation bar and the message box. It hides itself while you are inside the assistant, because the card is already in front of you.
• The keyboard does not cover the message box. When the on-screen keyboard opens, the conversation above squeezes up and the box and its Send button stay visible.
• Enter behaves as you set it. The return key on the on-screen keyboard shows Send or a new-line hint to match your Send messages with preference.
• A proposal card stacks on a narrow screen: field names sit above their values, and Approve, Edit and Deny stack with Approve first.
• The working block wraps: activity and elapsed time on the first line, Stop on the second, and it shows two recent steps instead of three.